猫史档案馆

大家好久不见呀

本次教程，我将教大家实现AI分类

本教程最终效果

 

在这里，我们需要使用brain.js作为依赖，它不太受环境约束，甚至可以在环境严格的浏览器下使用！

 

1. 导入

我们先简单的导入它:

创建brain.js文件，浏览器访问unpkg.com/brain.js（教程使用版本unpkg.com/brain.js@2.0.0-beta.20/dist/browser.js）将所有内容复制至brain.js文件，导入完成！

在index.js导包

const brain = require('./brain.js')

 

2. 创建数据

训练数据的格式为 用英文最容易训练

[
    {
      "input": "输入1",
      "output": "分类"
    },
    {
      "input": "输入2",
      "output": "分类"
    },
    ...
]

在这里，我要分类输入属于前端还是后端，便可以用如下数据 (当然，实际数据有20个，我会放文末）

[
    {
      "input": "the user interface component is fixed",
      "output": "frontend"
    },
    {
      "input": "mysql, mongo, firebase databases",
      "output": "backend"
    },
    {
      "input": "restful api is useful with backend",
      "output": "backend"
    },
    {
        "input": "data access layer is not presentation layer",
        "output": "backend"
    },
    {
        "input": "the web browser loads dynamic webpages slowly",
        "output": "frontend"
    }
  ]

(最后一个真实

将它们存入data常量方便使用

const data = [...]

注：如果数据太大，可以单独存个文件，应该都会

 

3. 训练模型

先确认使用的网络，这里我们用LS喵

初始化网络

const network = new brain.recurrent.LS喵()

创建训练函数，最好是异步

// 异步训练模型 (it为训练次数,不能过大)
async function train(it = 50) {
    world.say('Start Training')

    network.train(data, {
        iterations: it,
        log: true,
        logPeriod: 10,
        layers: [10],
    })

    world.say('Trained')
}

iterations: 训练次数，过大会被岛三制裁

log: 在控制台输出训练状态

logPeriod: 训练几次输出一次

layers: 中间层

network.train是同步执行函数，训练时会堵塞，注意使用时间！

之后在启动后调用即可

train()

// 也可以指定次数
// train(114)

还可以测试一下 将train()改为

train().then(()=>{
    console.log(network.run('User login'))
});
// 输出是对是错不重要，那是数据的问题，成功了才最重要

训练部分就完成了！

 

4. 使用

在训练的测试代码中就能知道 network.run() 就是使用函数，函数具体使用如下

network.run(value:string):string

value: 输入

返回: 分类

在这里我写了个简单的处理代码

world.onChat(({ message }) => {
    // $消息
    if (message.indexOf('$') == 0)
        world.say(network.run(message.slice(1)))

    // #指令 参数
    if (message.indexOf('#') == 0) {
        switch (message.split(' ')[0].slice(1)) {
            case 'train':
                train()
                break

            case 'data':
                const args = message.split(' ').slice(1);
                network.train([{ input: args[0], output: args[1] }])
                break;

            default:
                world.say('未知指令')
        }
    }
})

发送 $输入 即可～

本段代码健壮性不强，不建议直接使用

 

5. 其他

data:

[
    {
      "input": "the user interface component is fixed",
      "output": "frontend"
    },
    {
      "input": "the css file look inituitive",
      "output": "frontend"
    },
    {
      "input": "i need a few ui designs",
      "output": "frontend"
    },
    {
      "input": "the database has issues",
      "output": "backend"
    },
    {
      "input": "the button is centered",
      "output": "frontend"
    },
    {
      "input": "make it clickable",
      "output": "frontend"
    },
    {
      "input": "i did the api integration",
      "output": "backend"
    },
    {
      "input": "a driver code should have less memory usgae",
      "output": "backend"
    },
    {
      "input": "it needs more memory",
      "output": "backend"
    },
    {
      "input": "code with responsive design in users interface",
      "output": "frontend"
    },
    {
      "input": "navigate the website easily",
      "output": "frontend"
    },
    {
      "input": "user login and authentication",
      "output": "backend"
    },
    {
      "input": "forms and dropdowns lists",
      "output": "frontend"
    },
    {
      "input": "username password email are stored",
      "output": "backend"
    },
    {
      "input": "programming loading animation",
      "output": "frontend"
    },
    {
      "input": "mysql, mongo, firebase databases",
      "output": "backend"
    },
    {
      "input": "restful api is useful with backend",
      "output": "backend"
    },
    {
        "input": "data access layer is not presentation layer",
        "output": "backend"
    },
    {
        "input": "the web browser loads dynamic webpages slowly",
        "output": "frontend"
    }
  ]